Semantic anomaly detection method and system based on hybrid covariance and class-conditional calibration

Abstract

Disclosed is a semantic anomaly detection method and system based on hybrid covariance and class-conditional calibration, which belongs to the technical field of anomaly detection. Calculate the anomaly score of the colour image to be detected based on multiple network output feature values of the image to be detected, the class-conditional Mahalanobis distance, and the number of preset categories. The disclosed method and system can effectively realize anomaly detection.